ping yuweb.hku.hk/~pingyu/6066/slides/ln4_maximum theorem...the maximum theorem: the continuity of...
TRANSCRIPT
Ch04. Maximum Theorem, Implicit Function Theoremand Envelope Theorem
Ping Yu
Faculty of Business and EconomicsThe University of Hong Kong
Ping Yu (HKU) MIFE 1 / 27
1 The Maximum Theorem
2 The Implicit Function Theorem
3 The Envelope TheoremPing Yu (HKU) MIFE 2 / 27
Overview of This Chapter
The maximum theorem: the continuity of the optimizer and optimum with respectto (w.r.t.) a group of parameters.
The implicit function theorem: the differentiablity of the optimizer w.r.t. a group ofparameters.
The envelope theorem: the differentiablity of the optimum w.r.t. a group ofparameters.
In the previous two chapters, we are "moving along a curve", that is, f and G aregiven (or the parameter value is given) and we are searching over x along f to findthe optimal x�.
In this chapter, we are "shifting a curve", i.e., we change the parameter values toshift f and G, and check how the optimizer and optimum respond to such shifting.
Ping Yu (HKU) MIFE 2 / 27
The Maximum Theorem
The Maximum Theorem
Ping Yu (HKU) MIFE 3 / 27
The Maximum Theorem
History of the Maximum Theorem
Claude J. Berge (1926-2002), French
Ping Yu (HKU) MIFE 4 / 27
The Maximum Theorem
The Problem
Problem: We want to know how x� (or the maximized function) depends on theexogenous parameters a, rather than what x� is for a particular a.- For example, in consumer’s problem max
x1,x2u(x1,x2) s.t. p1x1+p2x2 = y , how
(x�1 ,x�2) or (u(x�1 ,x
�2)) depends on (p1,p2,y) rather than what (x�1 ,x
�2) is when
p1 = 2, p2 = 7, and y = 25, i.e., we are interested in what the demand function is.
Mathematically,max
x1,��� ,xnf (x1, � � � ,xn,a1, � � � ,ak )
s.t. g1(x1, � � � ,xn,a1, � � � ,ak ) = c1,...
gm(x1, � � � ,xn,a1, � � � ,ak ) = cm.
When is the "demand" function continuous?
Ping Yu (HKU) MIFE 5 / 27
The Maximum Theorem
Continuous Correspondence
Define the maximized value of f when the parameters are (a1, � � � ,ak ) as
v(a1, : : : ,ak ) = f (x�1(a1, : : : ,ak ), : : : ,x�n(a1, : : : ,ak ),a1, : : : ,ak ).
Using the feasible set, the problem can be restated as
maxx1,��� ,xn
f (x1, � � � ,xn,a1, � � � ,ak )
s.t. (x1, � � � ,xn) 2G(a1, � � � ,ak ),
where G(a1, � � � ,ak )� f(x1, � � � ,xn) j gj (x1, � � � ,xn,a1, � � � ,ak ) = cj , 8 jg is a setvalued function or a correspondence.
Two sets of vectors A and B are within ε of each other if for any vector x in one setthere is a vector x 0 in the other set such that x 0 2 Bε (x).
The correspondence G is continuous at (a1, : : : ,ak ) if 8 ε > 0, 9 δ > 0 such that if(a01, � � � ,a
0k ) is within δ of (a1, � � � ,ak ) then G(a01, � � � ,a
0k ) is within ε of
G(a1, � � � ,ak ). [Figure here]- If G is a function, then the continuity of G as a correspondence is equivalent tothe continuity as a function.- The continuity of the functions gj does not necessarily imply the continuity of thefeasible set. (Exercise)
Ping Yu (HKU) MIFE 6 / 27
The Maximum Theorem
Figure: Continuous and Discontinuous Correspondence
Ping Yu (HKU) MIFE 7 / 27
The Maximum Theorem
The Maximum Theorem
Theorem
Suppose that f (x1, � � � ,xn,a1, � � � ,ak ) is continuous (in (x1, � � � ,xn,a1, � � � ,ak )), thatG(a1, � � � ,ak ) is a continuous correspondence, and that for any (a1, � � � ,ak ) the setG(a1, � � � ,ak ) is compact. Then
(i) v(a1, � � � ,ak ) is continuous, and
(ii) if (x�1(a1, � � � ,ak ), � � � ,x�n(a1, � � � ,ak )) are (single valued) functions thenthey are also continuous.
Compactness of G(a1, � � � ,ak ) and continuity of f w.r.t. x guarantees the existenceof x� for any a by the Weierstrass Theorem.
Uniqueness of x� is guaranteed by the uniqueness theorem in Chapter 3.
Ping Yu (HKU) MIFE 8 / 27
The Implicit Function Theorem
The Implicit Function Theorem
Ping Yu (HKU) MIFE 9 / 27
The Implicit Function Theorem
History of the IFT
Augustin-Louis Cauchy (1789-1857), French
Ping Yu (HKU) MIFE 10 / 27
The Implicit Function Theorem
The Problem
Problem: How to solve x� 2Rn from n FOCs and how sensitive of x� is to theparameter?
Suppose that we have n endogenous variables x1, : : : ,xn, m exogenous variablesor parameters, b1, : : : ,bm, and n equations or equilibrium conditions
f1(x1, � � � ,xn,b1, � � � ,bm) = 0,f2(x1, � � � ,xn,b1, � � � ,bm) = 0,
...fn(x1, � � � ,xn,b1, � � � ,bm) = 0,
or, using vector notation,f(x,b) = 0,
where f : Rn+m !Rn, x 2Rn, b 2Rm, and 0 2Rn.
When can we solve this system to obtain functions giving each xi as a function ofb1, : : : ,bm?
Ping Yu (HKU) MIFE 11 / 27
The Implicit Function Theorem
The Case of Linear Functions
Suppose that our equations are
a11x1+ � � �+a1nxn+ c11b1+ c1mbm = 0
a21x1+ � � �+a2nxn+ c21b1+ c2mbm = 0...
an1x1+ � � �+annxn+ cn1b1+ cnmbm = 0
We can write this, in matrix notation, as
[A j C]�
xb
�= 0 or Ax +Cb = 0,
where A is an n�n matrix, C is an n�m matrix, x is an n�1 (column) vector, andb is an m�1 vector.
As long as A can be inverted or is of full rank,
x = �A�1Cb or∂x∂b0
= �A�1C.
Ping Yu (HKU) MIFE 12 / 27
The Implicit Function Theorem
The General Nonlinear Case
If there are some values (x,b) for which f (x,b) = 0, then a parallel result holds inthe neighborhood of (x,b) if we linearize f in that neighborhood.
Theorem
Suppose that f : Rn+m !Rn is a C1 function on an open set A�Rn+m and that (x,b)in A is such that f (x,b) = 0. Suppose also that
∂ f(x,b)∂x0
=
0BBB@∂ f1(x,b)
∂x1� � � ∂ f1(x,b)
∂xn...
. . ....
∂ fn(x,b)∂x1
� � � ∂ fn(x,b)∂xn
1CCCAis of full rank . Then there are open sets A1 �Rn and A2 �Rm with x 2 A1, b 2 A2 andA1�A2 � A such that for each b in A2 there is exactly one g(b) in A1 such thatf(g(b),b) = 0. Moreover, g : A2 ! A1 is a C1 function and�
∂g(b)∂b0
�n�m
= ��
∂ f(g(b),b)∂x0
��1
n�n
�∂ f(g(b),b)
∂b0
�n�m
.
Ping Yu (HKU) MIFE 13 / 27
The Implicit Function Theorem
Explanations of the IFT
Intuition: For (x,b) in a neighborhood of (x,b) such that f(x,b) = 0 with x = g(b),we have
∂ f(g(b),b)∂x0
∂g(b)∂b0
+∂ f(g(b),b)
∂b0= 0n�m,
so
∂g(b)∂b0
= ��
∂ f(g(b),b)∂x0
��1 �∂ f(g(b),b)
∂b0
���A�1C in the linear case.
This is a local result, rather than a global result.
If b is not close to b we may not be able to solve the system, and that for aparticular value of b there may be many values of x that solve the system, butthere is only one close to x. [Figure here]
For all values of b close to b̄ we can find a unique value of x close to x̄ such thatf (x ,b) = 0. However, (1) for each value of b there are other values of x far awayfrom x̄ that also satisfy f (x ,b) = 0, and (2) there are values of b, such as b̃ forwhich there are no values of x that satisfy f (x ,b) = 0.
That ∂ f (x̄ ,b̄)∂x is of full rank means jg0(x̄)j 6= 0 in this example.
Ping Yu (HKU) MIFE 14 / 27
The Implicit Function Theorem
Figure: Intuition for the IFT: f (x ,b) = g(x)�b, x ,b 2R
Ping Yu (HKU) MIFE 15 / 27
The Implicit Function Theorem
Comparative Statics
The IFT does not provide conditions to guarantee the existence of (x,b) such thatf (x,b) = 0; rather, it provides conditions such that if such an (x,b) exists, then wecan also uniquely solve f (x,b) = 0 in its neighborhood.
So the most important application of the IFT is to obtain ∂g(b)∂b0 rather than
guarantee the existence or uniqueness of the solution.
Example
The seller of a product pays a proportional tax at a flat rate θ 2 (0,1). Hence, theeffective price received by the seller is (1�θ )P, where P is the market price for thegood. Market supply and demand are given by the differentiable functions
Qd = D(P), with D0(�)< 0
Qs = S((1�θ )P), with S0(�)> 0
and equilibrium requires market clearing, that is, Qs = Qd . Analyze, graphically andanalytically, the effects of a decrease in the tax rate on the quantity transacted and theequilibrium price.
Ping Yu (HKU) MIFE 16 / 27
The Implicit Function Theorem
Solution to the Example
Market clearing requiresS((1�θ )P) = D(P). (1)
This equation implicitly defines the equilibrium price as a function P� = P (θ ) of theparameter θ .
Substituting the solution function P (�) back into (1), we have the identity
S [(1�θ )P (θ )] = D [P (θ )] .
Applying the IFT directly with f (P,θ ) = D [P (θ )]�S [(1�θ )P (θ )], we have
P 0 (θ ) =�PS0 (�)
D0 (�)� (1�θ )S0 (�) =(�)(�) > 0.
The quantity transacted in equilibrium is given by Q� = D [P (θ )], and therefore
dQ�
dθ= D0 (P�)P 0 (θ ) < 0.
Graphically, a reduction in the tax rate increases the effective price received by sellers forany given market price; these are therefore willing to sell any given quantity at a lowermarket price. Hence the supply curve shifts down. The equilibrium price falls, and theequilibrium quantity increases. [Figure here]
Ping Yu (HKU) MIFE 17 / 27
The Implicit Function Theorem
Quantity
Pric
e
Figure: Effect of a Tax Reduction
Ping Yu (HKU) MIFE 18 / 27
The Envelope Theorem
The Envelope Theorem
Ping Yu (HKU) MIFE 19 / 27
The Envelope Theorem
Intuition for The Envelope Theorem
Problem: How is v(a1, : : : ,ak ) sensitive to (a1, : : : ,ak )?
Suppose we are at a maximum (in an unconstrained problem) and we change thedata of the problem by a very small amount.
Now both the solution of the problem and the value at the maximum will change.
However at a maximum the function is flat (the first derivative is zero).
Thus when we want to know by how much the maximized value has changed itdoes not matter (very much) whether or not we take account of how the maximizerchanges or not.
In the figure in the next slide, f (x�(a0),a0)� f (x�(a),a0), i.e.,
∂v(a)∂a
=∂ f (x�(a),a)
∂x| {z }=0 by FOC
∂x�(a)∂a
+∂ f (x�(a),a)
∂a=
∂ f (x�(a),a)∂a
.
Ping Yu (HKU) MIFE 20 / 27
The Envelope Theorem
Figure: Intuition for the Envelope Theorem
Ping Yu (HKU) MIFE 21 / 27
The Envelope Theorem
SRAC and LRAC: A Motivating Example
The short run average cost (SRAC) of producing Q units when the scale is k begiven by SRAC(Q,k) and the long run average cost (LRAC) of producing Q unitsby LRAC(Q)..
Here, the scale of operation k , i.e., the size and number of plants and other fixedcapital, is assumed not to be changed for some level of production Q in the shortrun, but will be selected to be the optimal scale for that level of production in thelong run, so Q is like a and k is like x above.
In other words,LRAC(Q) =min
kSRAC(Q,k).
Let us denote, for a given value Q, the optimal level of k by k(Q).
Drawing one short run average cost curve for each of the (infinite) possible valuesof k . One way of thinking about the long run average cost curve is as the “bottom”or envelope of these short run average cost curves. [Figure here] This implies that
d LRAC(Q)dQ
=∂ SRAC(Q,k(Q))
∂Q.
Ping Yu (HKU) MIFE 22 / 27
The Envelope Theorem
Figure: The Relationship Between SRAC and LRAC
Ping Yu (HKU) MIFE 23 / 27
The Envelope Theorem
The General Problem
Consider again the maximization problem
maxx1,��� ,xn
f (x1, � � � ,xn,a1, � � � ,ak )
s.t. g1(x1, � � � ,xn,a1, � � � ,ak ) = c1,...
gm(x1, � � � ,xn,a1, � � � ,ak ) = cm.
The Lagrangian
L (x1, � � � ,xn,λ 1, � � � ,λ m;a1, � � � ,ak )
= f (x1, � � � ,xn,a1, � � � ,ak )+m
∑j=1
λ j (cj �gj (x1, � � � ,xn,a1, � � � ,ak )).
Suppose the optimal values of xi and λ j are x�i (a1, � � � ,ak ) and λ�j (a1, � � � ,ak ),
i = 1, � � � ,n, j = 1, � � � ,k . Then
v(a1, : : : ,ak ) = f (x�1(a1, : : : ,ak ), : : : ,x�n(a1, : : : ,ak ),a1, : : : ,ak ).
The envelope theorem says that the derivative of v is equal to the derivative of Lat the maximizing values of x and λ .
Ping Yu (HKU) MIFE 24 / 27
The Envelope Theorem
The Envelope Theorem
Theorem
If all functions are defined as above and the problem is such that the functions x� andλ� are well defined, then
∂v∂ah
(a1, � � � ,ak ) =∂L
∂ah(x�1(a1, � � � ,ak ), � � � ,x�n(a1, � � � ,ak ),
λ�1(a1, � � � ,ak ), � � � ,λ �m(a1, � � � ,ak );a1, � � � ,ak )
=∂ f
∂ah(x�1(a1, � � � ,ak ), � � � ,x�n(a1, � � � ,ak ),a1, � � � ,ak )
�m
∑j=1
λ�j (a1, � � � ,ak )
∂gh
∂ah(x�1(a1, � � � ,ak ), � � � ,x�n(a1, � � � ,ak ),a1, � � � ,ak )
for all h = 1, � � � ,k. In matrix and vector notation,
∂v∂a(a) =
∂L
∂a(x�(a),λ �(a);a)
=∂ f∂a(x�(a),a)� ∂g(x�(a),a)0
∂aλ�(a).
Ping Yu (HKU) MIFE 25 / 27
The Envelope Theorem
Interpretation of the Lagrange Multiplier
In Chapter 2, we have interpreted the Lagrange multiplier as the penalty onviolating the constraint.
More rigorously, treating cj , j = 1, � � � ,m, also as parameters, then by the envelopetheorem,
∂v∂cj(a1, � � � ,ak ,c1, � � � ,cm) = λ
�j (a1, � � � ,ak ,c1, � � � ,cm).
Intuition: Think of f as the profit function of a firm, the gj equation as the resourceconstraint, and cj as the amount of input j available to the firm. In this situation,∂v∂cj(a1, � � � ,ak ,c1, � � � ,cm) represents the change in the optimal profit resulting from
availability of one more unit of input j . Alternatively, it tells the maximum amountthe firm would be willing to pay to get another unit of input j . For this reason, λ
�j is
often called the internal value, or more frequently, the shadow price of input j . Itmay be a more important index to the firm than the external market price of input j .
Ping Yu (HKU) MIFE 26 / 27
The Envelope Theorem
Other Applications in Microeconomics
Consumer Theory:- Hotelling’s Theorem- Hicks-Slutsky equations- Roy’s Theorem- Shephard’s lemma
Production Theory:- Hotelling’s lemma- Shephard’s lemma (again!)
Ping Yu (HKU) MIFE 27 / 27